What 1,827 real ChatGPT prompts reveal about the coming agentic search hype

This guest post by Metehan Yesilyurt originally appeared on metehan.ai

I wanted to understand how people really use ChatGPT, so I did something simple. I searched Google for “site:chatgpt.com Temporary Chat” and checked Archive.org for similar URLs. These searches revealed thousands of ChatGPT conversation links. I downloaded all these URLs and extracted the q= parameters – these are the actual questions and prompts people typed into ChatGPT.

After decoding them, I had 1,827 real user queries to analyze.

This dataset is just a tiny glimpse into the billions of AI interactions happening every day. But even this small sample reveals something important: people aren’t using AI like a search engine anymore. They’re having conversations, giving commands, and asking AI to complete complex tasks.

The results confirm what many of us have suspected (and you already know it): we’re not just dealing with a smarter search engine — we’re witnessing the emergence of a new user behaviour paradigm.

This post explores data to provide insights and strategies for AI Search Optimization, GEO, AEO, or any other term you prefer.

The big picture: From keywords to conversations

The first thing that stands out from the data is the sheer complexity of user interactions. The era of the two- or three-word query is being supplemented by detailed, conversational prompts.

Important Note: 1827 queries are not showing the exact data correlation with overall ChatGPT prompts/chats. These queries are the only accessible data from the Google index & archive.org

Dashboard of Key Metrics:

Metric	Value	Insight
Total Queries Analyzed	1,827	A robust dataset illustrating emerging user behaviors.
Average Prompt Length (Words)	42	Users are providing detailed instructions, treating the AI as a collaborator, not just a search box.
Median Prompt Length (Words)	11	A mix of short, direct commands and long, detailed requests shows the AI’s versatile role.
Primary Intent	Task-Oriented	A staggering 75% of queries are not questions, but also commands to perform a specific action.

This data signals a fundamental shift.

Users aren’t just looking for a webpage — they’re looking for a result, an output, a completed task.

Deep dive: What are users actually doing?

A granular analysis of the query logs reveals fascinating patterns, particularly in the technical and commercial sectors.

AI as a co-pilot for developers

A massive 40% of all task-oriented queries were related to code and development. The AI is being used as an indispensable tool for debugging, learning, and increasing productivity.

Table 2.1: Breakdown of Technical & Developer Queries

Category	Sub-Category	Query Examples	Percentage of Technical Queries
Code Debugging	C++, Python	fix: class Solution { ... }, fix bool containsNearbyDuplicate(...)	35%
Code Explanation	Rust, JavaScript	what does this code do? let points = ..., explain this regex ...	25%
Code Conversion	Shell Scripting	could you please convert to fish shell lua?, convert to fish shell export REPOS=...	15%
Tooling & Config	Neovim, Docker	create a keymap to use jj to enter normal mode, run docker container with env file	15%
General Concepts	API, LSP	what are code actions lsp, C# throw	10%

The emergence of high-intent, hyper-local commercial queries

While this dataset is a tiny sample of a much larger global conversation, it reveals an interesting pattern: users are leveraging AI for high-intent, specific commercial inquiries.

Table 2.2: Analysis of Commercial Query Patterns

Intent Category	Industry/Product	Geographic Pattern	Query Examples from Dataset
B2B Procurement	Industrial Filters	City-Specific (e.g., Shanghai)	上海耐高温高效过滤器 (Shanghai high-temperature resistant high-efficiency filter)
B2B Procurement	Clean Room Equipment	City-Specific (e.g., Guangzhou)	风淋室价格 (air shower room price)
B2C Local Retail	Specialty Foods	City-Specific (e.g., Guangzhou)	广州五星级酒店月饼 (Guangzhou five-star hotel mooncakes)
Service Inquiry	SEO, Web Dev	International	海外推广方法,网站推广优化外链 (Overseas promotion methods, website promotion optimization external links)

The key insight here is not that Shanghai and Guangzhou are the epicenters of AI commerce, but that users globally are (also) likely performing similar, highly specific local searches. They expect the AI to understand not just “industrial filters,” but “high-temperature resistant filters available from a supplier in [my city].”

I don’t think 95% of websites are doing this optimization for local/wholesale/B2B. A huge money printing opportunity is here.

The art of the prompt: Users are learning to “program” AI

One of the most significant trends is the rise of sophisticated prompt engineering, particularly the use of “persona” prompts. Users are instructing the AI to “act as” an expert to frame the response.

Table 3.1: Common “Act As…” Persona Prompts

Requested Persona	Task Type	Example Prompt (Translated)
Food Critic	Creative Writing	“I want you to act as a food critic. I will tell you about a restaurant…”
Mental Health Advisor	Guidance/Consultation	“I want you to act as a mental health advisor. I will provide you with an individual…”
Time Travel Guide	Informational/Creative	“I want you to act as a time travel guide. I will provide the historical period…”
Stack Overflow Post	Technical Q&A	“I want you to act as a Stack Overflow post. I will ask programming questions…”
Recruiter	Professional Services	“I want you to act as a recruiter. I will give you information about job openings…”

This behaviour can be a game-changer.

It means that the AI’s response is not just based on the information it finds, but on the persona it adopts. For brands, this presents both a challenge and an opportunity: if you don’t define your brand’s persona for the AI, someone else will.

Please publish prompting guides for your brand, teach them. Otherwise, you’ll leave everything to the users.

Your new playbook: Actionable AEO, GEO, LLMO, AISO strategies for 2025 and beyond

Adapting to this new reality requires a shift in mindset and tactics. Below are concrete, actionable steps for different enterprise roles.

Table 4.1: For Technical SEOs & Developer Relations

Action Item	Rationale	Implementation Example
Structure Code for search engines	AI models are parsing code for “fix” and “explain” queries. Well-structured, commented code is more likely to be used as a definitive source.	Ensure code blocks are clean and accompanied by explanatory text.
Create “Convert To” Content	High volume of queries converting code between languages/frameworks (e.g., Bash to Fish).	Publish articles titled “How to Convert X to Y,” providing side-by-side code comparisons and explanations.
Build a Glossary of Errors	Users query specific error messages.	Create a knowledge base where each page targets a specific error code or message, providing a clear explanation and solution.
Optimize for Tool-Specific Queries	Users ask for specific configurations for tools like nvim, Docker, and eslint.	Develop “Cheatsheets” or “Configuration Guides” for popular developer tools in your ecosystem.

Table 4.2: For Content Strategists & Marketers

Action Item	Rationale	Implementation Example
Develop Brand Personas for AI	Users are assigning personas to the AI. Proactively define how your brand should be represented.	Create a public /ai-prompting-guide page on your website that outlines your brand’s tone, key messages, and preferred terminology for AI to use.
Optimize for Summarization	A large percentage of queries involve summarizing URLs. Content must be easily digestible by AI.	Use clear <h1>, <h2> structures. Start articles with a concise executive summary. Use bullet points for key features/benefits.
Create Workflow Automation Content	Users are automating multi-step tasks (e.g., analyze, verify, tweet).	Create content that facilitates these workflows. E.g., “A Marketer’s Guide to Automating Social Media Updates with AI.”
Target “Act As an Expert” Queries	Users seek expert-level input. Position your content as the source for that expertise.	Frame content around expert guidance: “An Expert’s Take on [Topic],” “A Financial Advisor’s Guide to [Topic].”

Table 4.3: For E-commerce & Local SEO Specialists

Action Item	Rationale	Implementation Example
Treat Product Data as an API	AI needs structured data to answer specific commercial queries (e.g., product specs, price, location).	Ensure all specifications (e.g., filter sizes, materials) are in machine-readable formats like tables.
Answer Hyper-Local Queries	Commercial queries are often tied to specific cities.	Create dedicated location pages that list specific products/services available in that city. Use local language and terminology. (not saying create spammy PSEO content)
Build Comparison Pages	Users will ask AI to compare products. Be the source for that comparison.	Publish in-depth “Product A vs. Product B” pages with detailed tables comparing features, specs, and pricing.
Include High-Intent Keywords	Queries include terms like “price,” “supplier,” “manufacturer,” and contact info.	Ensure your product and service pages explicitly include these terms and make contact information prominent and machine-readable.

Advanced strategies: Going beyond the basics

Creating AI-first content experiences

The future of content isn’t just about being visible. It’s also about being actionable. Here’s how to create content that AI can not only find but actively use:

1. Structured Tutorials with Checkpoints: Break down complex processes into discrete, verifiable steps that AI can track and validate.
2. Interactive Calculators and Tools: Develop web-based tools that AI can invoke to perform calculations or generate results for users.
3. API Documentation 2.0: Move beyond static docs to interactive API explorers that AI can use to test and demonstrate functionality.
4. Conversational Content Paths: Design content that anticipates follow-up questions and provides clear pathways to related information.

Preparing for multi-modal AI interactions

As AI becomes more sophisticated, queries will increasingly include images, voice, and other media:

1. Image SEO for AI: Ensure all images have descriptive alt text, captions, and surrounding context that AI can parse.
2. Video Transcription and Chaptering: Make video content AI-accessible with accurate transcriptions and time-stamped chapters.
3. Voice-Optimized Content: Structure content to be easily read aloud and understood in audio format.

Building AI-friendly information architecture

Your site’s structure needs to facilitate AI navigation:

1. Clear Taxonomies: Develop logical, hierarchical categorization that AI can understand and navigate.
2. Semantic URLs: Use descriptive, keyword-rich URLs that convey content meaning.
3. Comprehensive Internal Linking: Create rich internal link networks that help AI understand content relationships.
4. Dynamic Sitemaps: Maintain updated sitemaps that include metadata about content type, update frequency, and priority.

Measuring Success in the AI Era

Traditional SEO metrics need evolution for AISO:

New metrics to track:

1. AI Citation Rate: How often AI systems reference your content in responses
2. Task Completion Rate: Percentage of user tasks successfully completed using your content
3. Semantic Coverage Score: How comprehensively your content covers related concepts
4. Query Resolution Depth: Average number of your pages AI needs to fully answer queries
5. Brand Persona Accuracy: How accurately AI represents your brand voice and values

Tools and techniques for monitoring

1. AI Query Logs: Analyze logs from AI platforms (where available) to understand how your content is being used
2. Synthetic Monitoring: Regularly test how AI systems respond to queries about your products/services
3. Competitive Intelligence: Monitor how AI represents competitors to identify gaps and opportunities
4. User Feedback Loops: Implement systems to collect feedback on AI-mediated interactions with your content

The future is agentic, task-oriented, context-aware and conversational

The data from our analysis is clear: the way users interact with information online is undergoing a radical transformation. Even though it’s a tiny dataset, there are no traditional patterns. The line between search, content creation, and task automation is blurring. For SEOs, marketers, and business leaders, the challenge is to move beyond optimizing for keywords and start optimizing for outcomes.

The winners in this field will be those who:

• Structure their data for AI consumption
• Create content that facilitates task completion
• Build systems that can provide real-time, contextual information
• Develop clear brand personas for AI interactions
• Measure and optimize for new engagement patterns

By developing these changes and implementing the strategies outlined above, we can not only survive but thrive in this new/updated, exciting field. The future of search is here, and it’s asking us to do more than just provide answers.

It’s asking us to help get the job done.